Meta's Muse AI agent surges to top app charts amid security fixes

Internal documents from a lawsuit against OpenAI and Microsoft reveal top executives privately acknowledging their AI systems were built on what one described as potentially the largest theft of labor in human history. The filings, unsealed in a case brought by publishers including Mother Jones, show Microsoft director of applied science Brent Hecht warning of a coming "doom loop" where AI training data scraped from the web would erode the economic base of the very sites and creators supplying that content. Hecht noted the process threatens essential suppliers and risks making the broader internet unusable over time.

These admissions align with ongoing claims from multiple news organizations that companies like OpenAI and Microsoft trained models on copyrighted material without permission or compensation. The publishers argue this amounts to systematic appropriation of years of reporting and creative work to power commercial products now valued in the hundreds of billions. Microsoft and OpenAI have defended the practice as fair use, yet the internal comments suggest awareness that the scale of the copying far exceeds traditional legal boundaries.

The same pattern of aggressive data use appears at Meta, where chief AI officer Alexandr Wang has promoted the company's new Muse agent as a tool that can instantly save users around a thousand dollars through automated tasks like canceling subscriptions or negotiating lower insurance rates. Early users reported modest wins, such as recovering payments for unused services or switching policies. Wang framed the rollout as a direct benefit to ordinary people, encouraging public posts under a challenge hashtag.

Yet security researchers quickly identified serious problems with Muse on macOS. A zero-day vulnerability allowed local attackers to redirect the agent's cloud processing to their own servers, gaining access to user accounts and potentially executing commands without clear alerts. The flaw stemmed from design choices including cloud-based dictation and broad permissions granted to other applications. Meta issued a patch shortly after the issue surfaced but maintained that real-world risks remained low since the exploit required physical device access.

Critics see these developments as part of a larger pattern where Silicon Valley prioritizes rapid deployment over safeguards or respect for existing property rights. The labor theft documented in the OpenAI case involves the foundational training data that makes tools like Muse possible in the first place. Meanwhile, the security shortcomings indicate that even after launch, these systems receive less rigorous protection than traditional software. Ordinary users and creators bear the costs while executives chase valuation growth.

The lawsuits continue to move forward, with similar actions filed by major outlets like The New York Times. Internal warnings from within the companies themselves suggest the current approach may prove unsustainable, as the supply of quality data dries up and legal challenges mount. For now, the push to embed AI agents deeper into daily life proceeds despite these unresolved questions.